Document 10677340

advertisement
c
Applied Mathematics E-Notes, 8(2008), 36-39 Available free at mirror sites of http://www.math.nthu.edu.tw/∼amen/
ISSN 1607-2510
A Weak Regular Splitting For Singular Linear
Systems∗
Guang-Hui Cheng and Ting-Zhu Huang†
Received 23 November 2006
Abstract
In this paper, the semiconvergence of a proper weak regular splitting method
for the singular linear algebraic system Ax = b with a singular M -matrix A is
discussed. The extrapolated iterative method based on the weak regular splitting
iterative method is also constructed.
Let us consider the solution of a linear system of equations
Ax = b
(1)
by iterative methods. Here A is an n × n singular matrix, x, b ∈ Rn with b known and
x unknown. We assume that the system (1) is solvable, i.e., it has at least a solution,
which is equivalent to b ∈ R(A), where R(A) denotes the range of A.
In order to solve the system (1) with iterative methods usually we construct the
iteration of the form
x0 ∈ Rn;
xk+1 = T xk + c,
k = 0, 1, 2, ... .
(2)
Here the question is the construction of the iterative matrix T and the vector c. The
usual approach is through the splitting of the coefficient matrix A:
A = M − N,
by the formulas
T = M −1 N,
c = M −1 b,
where M is nonsingular.
It is well known that for singular system the iterative method (2) is semiconvergent
if, and only if, the associated convergence factor
υ(T ) ≡ max{|λ|, λ ∈ σ(T )\{1}} < 1,
(3)
and the elementary divisors associated with µ = 1 ∈ σ(T ) are linear, i.e.,
index(I − T ) = 1,
∗ Mathematics
(4)
Subject Classifications: 65F10
of Applied Mathematics, University of Electronic Science and Technology of China,
Chengdu, Sichuan 610054, P. R. China
† School
36
G. H. Cheng and T. Z. Huang
37
where υ(T ) denotes the pesudo-spectral radius of T , σ(T ) denotes the spectrum of T
and index(B) denotes the index of the matrix B, i.e., the smallest nonnegative integer
k such that rank(B k ) = rank(B k+1 ). In this case, T is called a semiconvergent matrix.
DEFINITION 1 [1]. A matrix A = (aij ) ∈ Rn×n is called a singular M -matrix if A
can be expressed in the form
A = sI − B,
s > 0,
B > 0,
(5)
and s = ρ(B), where ρ(B) denotes the spectral radius of B.
A singular M -matrix A is said to have “property c” (cf. [1]) if it can be splitted
into (5) and the matrix T = B/s is semi-convergent.
DEFINITION 2 [1]. Let A ∈ Rn×n. The splitting A = M −N is called weak regular
if M −1 > 0 and M −1N > 0.
In [4], the proper weak regular splitting for nonsingular M -matrix was constructed.
In the following we prove that for any irreducible singular M -matrix A ∈ Rn×n, there
exists an efficient weak regular splitting.
Without loss of generality we can partition A into
a
A1
A = (aij ) =
,
bT ann
where A1 is a nonsingular M -matrix of order n − 1, a 6 0, b 6 0 are in Rn−1 and
ann = bT A−1
1 a. There exist lower and upper triangular nonsingular M -matrices L1
and U1 , respectively, such that
A1 = L1 U1 .
Set
L = (lij ) =
L1
bT U1−1
0
1
and U = (uij ) =
U1
0
L−1
1 a
0
.
We need the following lemma.
LEMMA 1. Let A be an irreducible singular M -matrix and the splitting A = M −N
be a weak regular. Then it holds
ρ(M −1 N ) = 1,
index(I − M −1 N ) = 1.
Indeed, there exists a vector x > 0 such that Ax > 0 since A is an irreducible
singular M -matrix. By [3, Theorem 6], the result is proved.
Now, we set
Λ = diag(λ1 , λ2, ..., λn), 0 < λi < 1, i = 1, 2, ..., n − 1, λn > 0
and introduce the notation
D = diag(d1 , d2, ..., dn),
where
di =
1 − λi
> 0,
uii
i = 1, ..., n − 1,
d n = λn .
38
Singular Linear Systems
THEOREM 1. The matrices
M = LD−1 ,
N = M −A
(6)
define a weak regular splitting of the matrix A.
PROOF. Since L is a nonsingular M -matrix, M −1 = DL−1 > 0. N = L(D−1 − U ).
Therefore we have
M −1N = DL−1 L(D−1 − U ) = I − DU.
The matrix DU is an upper triangular singular M -matrix with the diagonal entries
1 − λi , i = 1, 2, ..., n − 1, and λn = 0, which proves the Theorem.
THEOREM 2. Assume that the numbers λi satisfy
0 < λi < 1,
i = 1, 2, ..., n − 1, λn > 0.
Then the iterative method induced by the weak regular splitting of the form (6) is
semiconvergent.
PROOF. By the proof of Theorem 1 we know the diagonal entries of the matrix
DU are 1−λi (i = 1, 2, · · · , n−1) and 0. Clearly, all eigenvalues of the iterative matrix
M −1N are λi (i = 1, 2, · · · , n − 1) and 1. Therefore υ(T ) = max |λi | < 1. By Lemma
1, M −1N is semiconvergent. The proof is complete.
The extrapolated iterative method based on the stationary iterative method (2) is
as follows:
x0 ∈ Rn;
xk+1 = (1 − ω)xk + ω(T xk + c),
k = 0, 1, 2, ... ,
(7)
where the parameter ω ∈ R\{0}.
Several papers (cf. [5,6,7]) studied the extrapolated iterative methods for singular
linear systems. In the following we study the extrapolated iterative methods for singular
linear systems base on the weak regular splitting iterative method (6).
THEOREM 3. The extrapolated iterative method (7) based on the weak regular
splitting iterative method (6) converges if, and only if, the following conditions are
satisfied
(a) index(I − T ) = 1;
2
(b) 0 < λi < 1 , λn > 0, and 0 < ω < 1−min
, (i = 1, 2, ..., n − 1).
λi
PROOF. Denote the extrapolated iteration matrix of (7) by
Tω = ωT + (1 − ω)I = I − ω(I − T ) = I − ωDU.
(8)
Then
I − Tω = ω(I − T ),
from which we have, obviously,
index(I − Tω ) = index(I − T ).
(9)
The equality (8) implies
ν(Tω ) = max |1 − ω(1 − λi )|,
i = 1, 2, ..., n − 1.
(10)
G. H. Cheng and T. Z. Huang
39
From (10) we deduce that |1 − ω(1 − λi )| < 1 if, and only if,
0<ω<
2
,
1 − min λi
i = 1, 2, ..., n − 1.
(11)
According to (9) and (11), Theorem 3 follows due to (3) and (4). The proof is complete.
Acknowledgment. Supported by NCET of Chinese ministry of education (NCET04-0893), scientific and technological key project of Chinese ministry of education
(107098), Sichuan province foundation for applied basic research and leader of science
and technology (05JY029-068-2).
References
[1] A. Berman and R. J. Plemmons, Nonnegative Matrices in the Mathematical Sciences, Academic, New York, 1979.
[2] D. M. Young, Iterative Solution of Large Linear Systems, Academic Press, New
York, 1971.
[3] M. Neumann and R. J. Plemmons, Convergent nonnegative matrices and iterative
methods for consistent linear systems, Numer. Math., 31(1978), 265–279.
[4] István Faragó, Proper weak regular splitting for M -matrices, Numerical analysis
and its applications (Rousse, 2000), 285–291, Lecture Notes in Comput. Sci., 1988,
Springer, Berlin, 2001.
[5] Y. Z. Song, Semiconvergence of extrapolated iterative methods for singular linear
systems, J. Comput. Appl. Math., 106(1999), 117–129.
[6] Y. Z. Song and L. Wang, On the semiconvergence of extrapolated iterative methods
for singular linear systems, Appl. Numer. Math., 44(2003), 401–413.
[7] Z. H. Cao, Semiconvergence of extrapolated iterative method for singular linear
systems, Appl. Numer. Math., 156(2004), 131–136.
Download